• No results found

A Survey on CBIR using Low Level Feature Combination

N/A
N/A
Protected

Academic year: 2020

Share "A Survey on CBIR using Low Level Feature Combination"

Copied!
10
0
0

Loading.... (view fulltext now)

Full text

(1)

International Journal of Emerging Technology and Advanced Engineering

Website: www.ijetae.com (ISSN 2250-2459,ISO 9001:2008 Certified Journal, Volume 4, Issue 11, November 2014)

211

A Survey on CBIR using Low Level Feature Combination

Chintan K. Panchal

1

, Risha A. Tiwari

2

1,2Computer Engineering Department, Hasmukh Goswami College of Engineering, Vahelal, Ahmadabad, India

Abstract—In the field of Image Processing, Image Retrieval has become one of the most active research areas in the past few years. In the Image Retrieval three main categories are there, text-based, content-based and semantic-based. In Content based Image Retrieval (CBIR), images are retrieved by their visual content such as color, texture and shapes. CBIR is used to effectively retrieve required images from large collection of images. In this paper the techniques of content based image retrieval are discussed, analyzed and compared. There are some different techniques for extracting color, texture and shape features.

KeywordsCBIR, Feature extraction,

I. INTRODUCTION

Image retrieval is technique concerned with searching and browsing digital images from database collection. This area of research is very active research since the 1970s [1]. Due to more and more images have been generated in digital form around the world, image retrieval attracts interest among researchers in the fields of image processing, multimedia, digital libraries, remote sensing, astronomy, database applications and other related area [1]. Fast retrieval of images has not always been easy, especially when you are working with thousands of images. For effectiveness of image retrieval system, it needs to operate on the collection of images to retrieve the relevant images based on the query image which conforms as closely as possible to human perception.

The purpose of an image database is to store and retrieve an image or image sequences that are relevant to a query. The database provided by James S. Wang is used for this purpose. In the WANG database 1,000 and 10,000 images has been used. In WANG database, the images are divided into 10 classes [2].Each class contain 100 images and 1,000 images has been used respectively. Classification of the images in the database into 10 classes makes the evaluation of the system easy.

In image retrieval research, researchers are moving from keyword based, to content based then towards semantic based image retrieval and the main problem encountered in the content based image retrieval research is the semantic gap between the low-level feature representing and high-level semantics in the images [1].

A.Keyword Based Image Retrieval

In the 1970s, the Keyword Based Image Retrieval system used keyword as descriptors to index an image [1]. In Fig.1General Framework of keyword based image retrieval is show.

Fig.1 General Framework of Keyword Based Image Retrieval [1].

In this technique when images are stored in the database they are examined manually and assigned keywords that are most appropriate to describe their contents. The Keyword which stored in the database, are stored as the part of the attributes associated to the image. During query stage, the image retrieval system will accept from the user one or many keywords which constitute the search criteria. A keyword matching process is then performed to retrieve images associated with the keywords that match the search criteria.

B.Content Based Image Retrieval

(2)

International Journal of Emerging Technology and Advanced Engineering

Website: www.ijetae.com (ISSN 2250-2459,ISO 9001:2008 Certified Journal, Volume 4, Issue 11, November 2014)

212

Fig.2General Framework of Content Based Image Retrieval [1]

II. FEATURE EXTRACTION TECHNIQUES

Feature Extraction Techniques may include both text based features and visual features. Within CBIR visual feature are required to extract. In the visual features scope it can be classified as low level and high level features. The selection of the features to represent an image is one of the keys of a CBIR system. Because of perception subjectivity and the complex composition of visual data, there does not exist a single best representation for any given visual feature [4]. Multiple approaches have been introduced for each of these visual features and each of them characterizes the feature from a different perspective [4].Main three low level features are the Color, Texture and Shape.

A. Color Feature

Color features are the basic characteristics for the content of images. With the color feature human can identifies and distinguish between object and images. Colors are used in image retrieval because they are powerful descriptors and sometimes provide powerful information about images. To extract the color features from the content of an image, we need to select a color space and use its properties in the extraction. In common, colors are defined in three dimensional color spaces. In digital image purposes, RGB color space is the most prevalent choice [2]. The main drawback of the RGB color space is that it is perceptually non-uniform and device dependent system. The HSV color space is an intuitive system, which describes a specific color by its hue, saturation, and brightness values [2]. This color system is very useful in interactive color selection and manipulation [2].

For the description of color feature many methods can be used. They are Color Histogram, Conventional Color Histogram, Invariant Color Histogram, Fuzzy Color Histogram, Geometric Moment, Average RGB, Color Moment, Color Correlogram, and Color Coherence Vector.

I. Color Hitogram

A color histogram [5] represents the distribution of colors in an image, where each histogram bin corresponds to a color in the quantized color space. Color histograms are a set of bins where each bin represents a particular color of the color space being used. The number of bins depends on the number of colors there in the image. A color histogram for a given image is defined as a vector [5]:

H={H[1],H[2],H[3],H[4],…….H[i],……H[n]} Where i represents the color bin in the color histogram and H[i] represents the number of pixels of color i in the image, and n is the total number of bins used in the color histogram [5]. Every pixel in an image must be assigning a bin of a color histogram of that image. In order to compare images of different sizes, color histograms should be normalized [5]. The normalized color histogram is defined as: H'

H' = {H'[1],H'[2],H'[3],H'[4],……H'[i],……H'[n]} Where, H'= H[i]/p , p is the total number of pixels of image. However, color histogram has its own drawbacks. If two images have exactly the same color proportion but the colors are scattered differently, then we can’t retrieve correct images [5].

II. Conventional Color Histogram

(3)

International Journal of Emerging Technology and Advanced Engineering

Website: www.ijetae.com (ISSN 2250-2459,ISO 9001:2008 Certified Journal, Volume 4, Issue 11, November 2014)

213 III. Invariant Color Histogram

Color histograms have been widely used for object recognition. Though in practice these histograms often vary slowly under changes in viewpoint, it is clear that the color histogram generated from an image surface is intimately tied up with the geometry of that surface, and the viewing position [6].The Invariant Color Histogram is developed to create a color histogram with using color gradient and its invariant under any mapping of the surface which is locally

affine and so that a very wide class of viewpoint changes

IV. Fuzzy Color Histogram

A color Histogram is created by using a color space by dividing into a number of bins and after that by counting the number of pixels of the image that belong to each bin. In the result similar color problem is appearing. Similar images have small differences in scene or contain noise will produce histograms with dissimilar adjacent bins and vice versa. In order to present a solution to this problem, the FCH uses a small number of bins produced by linking the triplet from the L*a*b* color space into a single histogram by means of a fuzzy expert system. The a* and b* components are considered to have more weight than L* as it is mostly the combination of the two which provides the color information of an image [6]. The L*a*b* color space was selected because it is a perceptually uniform color space which approximates the way that humans perceive color. In L*a*b*, L* stands for luminance, a* represents relative greenness-redness and b* represents relative blueness-yellowness [6]. All colors and grey levels can be expressed throughout a combination of the three components. However, L* does not contribute in providing any unique color but for shades of colors, white, black and grey. Thus, the L* component receives a lower weight with respect to the other two components of the triplet [6].

V.Geometric Moments

The Geometric moment also called Image moment of an image. After the object segmentation image moment is useful to describe objects. An image moment is ascertain particular weighted average (moment) of the image pixels’ intensities [7]. With image moments Simple properties of the image are found, include area (or total intensity), its centroid, and information about its orientation. This feature use only one value for the feature vector, however, the performance of current implementation isn’t well scaled, which means that when the image size becomes relatively large, computation of the feature vector takes a large amount of time [7].

VI. Average RGB

The objective of use this feature is to filter out images with larger distance at first stage when multiple feature queries are involved [7]. It uses a small number of data to represent the feature vector and it also uses less computation as compared to others that’s why this feature is used. However, the accuracies of query result could be significantly impact if this feature is not combined with other features [7].

VII. Color Moments

Color moments are used as feature vectors for image retrieval to overcome the quantization drawback of the color histogram. The image is divided horizontally into three equal non-overlapping regions and from each of the three regions, we extract from each color channel the first three moments of the color distribution and store the 27 floating point numbers in the index of the image [7]. The first-order (mean), the second (standard deviation), and the third-order (skew-ness) color moments have been proved to be efficient and effective in representing color distributions of images, If the value of the ith color channel at the j th image pixel is pi j , then the color moments are as follows [2]:

Moment 1: Mean

1

1

N

i j ij

E

p

N

Moment 2: Standard Deviation

2

1

1

N

i j

p

ij

E

i

N

Moment 3: Skew-ness

3

1

1

N

s

i j ij i

s

p

E

N

VIII.Color Correlogram

(4)

International Journal of Emerging Technology and Advanced Engineering

Website: www.ijetae.com (ISSN 2250-2459,ISO 9001:2008 Certified Journal, Volume 4, Issue 11, November 2014)

214 IX. Color Coherence Vector

A color coherence vector is a split histogram which partitions pixels according to their spatial coherence [4].According to image pixel whether it is part of a larger region of uniform color, each pixel within the image is partitioned into two types, i.e., coherent or incoherent. Separate histograms can then be produced for both coherent and incoherent pixels thereby including some spatial information in the feature vector [4]. Because of the additional spatial information, it has been shown that CCV provides better retrieval results than the color histogram, mostly for those images which have either uniform color or texture regions. The HSV color space provides better results than CIE L*u*v* and CIE L*a*b* space for both the color histogram and color coherence vector representation.

B. Texture Feature

There is no formal definition for texture, but it can say that it provides the measure of properties such as smoothness, coarseness, and regularity. In addition, texture can be expressed as repeated patterns of pixels over a spatial domain. If the texture has exposed to some noise, the patterns and their repetition in the texture can be random and unstructured [2]. Because of there is no formal mathematical definition for texture, many different methods are proposed for computing texture but among those methods, no single method works best with all types of texture. Some common methods are used for texture feature extraction such as Discrete Wavelet Transform, Gabor Wavelet Transform, Haar Discrete Wavelet Transforms, Ranklet Transform, Fourier Transform, discrete cosine transform, Hadamard Transform, Gaussian Pyramid, Laplacian Pyramid, Steerable Pyramid, Gabor Filter.

I. Discrete Wavelet Transform

For analyze images in a multi-scale framework the two dimensional discrete wavelet transform (DWT) is an effective tool.It is also an effective tool to capture localized image details in both space and frequency domains. Through the Mallat’s tree algorithm the DWT is efficiently implemented. On original image DWT applies iterative linear filtering and critical down-sampling that yielding three high-frequency directional sub-bands at each scale level, in addition to one low-frequency sub-band usually known as image approximation. Directional sub-bands are sparse sub-images exhibiting image details according to horizontal, vertical and diagonal orientations [8].

[image:4.612.330.551.228.490.2]

The decomposition process is illustrated in Fig.3 with the top image undergoing a first level decomposition to generate 3 detail sub-bands (H1, V1, and D1), and one image approximation (A1). At the second level of decomposition, the approximation image (A1) undergoes the same process to produce a second scale level of image details (V2, H2 and D2) and a new image approximation (A2) [8]. The resulting 2-level DWT is shown in Fig. 4.

Fig. 3. A two-level wavelet decomposition [8].

Fig. 4. Resulting sub-bands for a 2-level DWT (original image in Fig. 3.) [8].

II.Gabor Wavelet Transform

Gabor wavelet transform (GWT) is an important classical method for multichannel, multi-resolution analysis of the image.GWT is represents image variations at different scales. Gabor wavelet transform provides a flexible method for designing efficient algorithms to capture more orientation and scale information [8]. Many researches stated that Gabor wavelet transform represents one of the efficient techniques for image texture retrieval yielding good results in content-based image retrieval applications due to many reasons [8]:

(5)

International Journal of Emerging Technology and Advanced Engineering

Website: www.ijetae.com (ISSN 2250-2459,ISO 9001:2008 Certified Journal, Volume 4, Issue 11, November 2014)

215

- Offering the capacity for edge and straight line detection with variable orientations and scales.

[image:5.612.52.289.178.320.2]

- Not being sensitive to lighting conditions of the image.

Fig. 5. A set of real impulse responses: multiscale, multi-orientation Gabor wavelet filters [8]

Drawback of Gabor wavelet representations is, it suffer from some weaknesses such as their costly computational complexity. GWT has non invariance to rotation as well as the non orthogonal property of the Gabor filters that implies redundancy in the filtered images.

III. Haar Discrete Wavelet Transforms

The wavelet transform represents a function as a superposition of a family of basis functions called wavelets [9]. Wavelet transforms extract information from signal at different scales by passing the signal through low pass and high pass filters [9].After the invention by HAAR, Haar Wavelet are being widely used. Haar wavelets are used to compute feature signatures, due to their fastest to computation and also have been found to perform well in practice. With Haar wavelets, it enables to speed up the wavelet computation phase for thousands of sliding windows of varying sizes in an image. The Haar wavelet's mother wavelet function can be described as [9]:

 

1 1, 0

2 1

1, 1

2 0,

t

t t

otherwise

 

 

  

and its scaling function _(t) can be described as:

 

1, 0

1

0,

t

t

otherwise

 

IV. Fourier Transform

Fourier Transform has been the most important part of signal transform. Fourier Transform converts a signal from the time domain into the frequency domain to measure the frequency components of the signal and For CBIR, Fourier Transform was used to extract texture features from high-frequency components of the image. Unfortunately, Fourier Transform failed to capture the information about objects locations in an image and could not provide local image features [11]. The Fourier Transform is an important image processing tools which is used to decompose an image into its sine and cosine components [10]. While the input image is in the spatial domain equivalent, the output of the transformation represents the image in the Fourier or frequency domain. In the Fourier domain image, each point represents a particular frequency contained in the spatial domain image [10]. The practical implementation of the DFT always uses the Fast Fourier transform (FFT). FFT is an algorithm that computes the discrete Fourier transform much faster than other algorithms. FFT is surely the most widely used signal processing algorithm and is the basic building block for a large percentage of algorithms in current usage and the 2D FFT pair is given by [10]:

1 1 2 ( )

0 0

1

[ , ]

[ , ]

mk ni

N M j

M N n m

f k l

F m n e

MN

 

 



1 1 2 ( )

0 0

1

[ , ]

[ , ]

mk ni

N M j

M N

n m

f m n

F k l e

MN

 

 



Where, 0 m, k M-1,0 n,1 N-1.

V.Ranklet Transform

(6)

International Journal of Emerging Technology and Advanced Engineering

Website: www.ijetae.com (ISSN 2250-2459,ISO 9001:2008 Certified Journal, Volume 4, Issue 11, November 2014)

216

This means that for an image, vertical, horizontal, and diagonal ranklet coefficients are computed. Finally, it is multi-resolution that the Ranklet Transform can be calculated at different resolutions using Haar wavelet supports [11].

VI. Discrete Cosine Transform

Discrete Cosine Transform (DCT) is used because of it is an important functions in many signal and image processing applications. At the initial stage of compression of all JPEG images use the discrete cosine transform. Just as the Fourier transform uses sines and cosines waves to represent a signal, the DCT uses N real basis vectors whose components are cosines [12]. It is shown that DCT can not only be used for compressing images, it can also be conveniently used for Content Based Image Retrieval [12].

VII. Hadamard Transform

The Hadamard Transform is much easier and quicker than sinusoidal transforms, since no multiplications are required. The Hadamard transform is based on the Hadamard matrix which a square array is having entries of 1 and – 1.The Hadamard matrix of order 2 is given by [12]

Hadamard matrices of order 2n can be recursively generated using the kronecker product

H(2n ) =H(2) × H (2n-1) If n=1, H(2) =H (2) n=2, H(4)=H(2) × H(2)

If n=3

Similarly, If n=7

If f is a N ×N image and H (N) is a N ×N transformation matrix then Hadamard transform is given by

VIII.Gaussian Pyramid

To extract features and remove noise form image, the Gaussian pyramid is used. It is also used to decompose images into information at multiple scales. The Gaussian pyramid consists of low-pass filtered, down-sampled version of the previous level of the pyramid, where the base level is defined as the original image. Fig. 6 shows the decomposition of an image into its Gaussian Pyramid representation [12].

Fig.6. Gaussian pyramid decomposition[12].

IX. Laplacian Pyramid

The Laplacian Pyramid is describe as the decomposition of the original image into a hierarchy of images. Fig. 7 shows the decomposition of an image into its Laplacian Pyramid representation. The original image is at the upper left corner. The images immediately below and to the right of the original image are the coarse and detail signal respectively resulting from the first level of decomposition of the original image. The images adjacent to and below the coarse signal of the first level of decomposition are the detail and coarse signals respectively of the second level of decomposition [12].

(7)

International Journal of Emerging Technology and Advanced Engineering

Website: www.ijetae.com (ISSN 2250-2459,ISO 9001:2008 Certified Journal, Volume 4, Issue 11, November 2014)

217 X.Steerable Pyramid

With the Steerable Pyramid, it can generate a multi-scale and multi-directional representation of the image. In The Steerable Pyramid image is decomposed into low-pass sub-band and high-pass sub-sub-band. In the low-pass sub-sub-band decomposition is iterated. The Steerable Pyramid decomposition is similar to the two-dimensional discrete wavelet transform but with directional sub-bands [12].

Steps to extract features from a color image by using Steerable Pyramid are as follows:

 This method first resizes the image to 128 ×128. Then divide the image into R, G, and B components.

 Apply the low pass filter and high pass filter on each component (R, G, and B).

 The low frequency content is the most important part of the signal. High frequency component contains less information as compared to low frequency component. Hence, the output of the low pass filter received from the first stage can be down sampled by a factor of 2. The process of filtering and down sampling can be repeated to get a multi-level decomposition.

 Then directional sub-bands (8 Nos.) are obtained from the output of low pass filter of each stage.

 Compute the features such as mean and standard deviation of directional sub-bands of query image as well as images in the database. Then Normalized Euclidean Distance is used to compute the similarity measure between query image and images from the database.

XI. Gabor Filter

For the Texture Features extraction most widely use statistical method is the Gabor filter. This is most usable method for the Texture. With Gabor filter are many approaches proposed to characterize textures of images. In most of the CBIR systems based in Gabor wavelet, the mean and standard deviation of the distribution of the wavelet transform coefficients are used to construct the feature vector and the basic idea of using Gabor filters to extract texture features is as follows [13]. A two dimensional Gabor function g(x, y) is defined as:

 

22 22

1

1

,

exp

2

2

x y

2

x y x

x

y

g x y

jw

 

 

 

 

 

Where

x are

y the standard deviations of the

Gaussian envelopes along the x and y direction. Given and image I(x,y) it’s Gabor transform is defined as

*

1 1 1 1

( , )

( , )

,

mn mn

w

x y

I x y g

x

x y

y dx dy

Where * indicates the complex conjugate. Then the mean

mn and the standard deviation

mn of the

magnitude of

w

mn

( , )

x y

The texture similarity measurement of a query image Q and a target image T in the database is defined by

( , )

m n mn

( , )

d Q T

 

d

Q T

Where

(

mnQ mnT

)

(

mnQ Tmn

)

mn Q T Q T

mn mn mn mn

d

If 00, 00, 01, 01,... 35, 35,

Q g

f

 

   

 

denote texture feature vector of query image And

00, 00, 01, 01,... 35, 35,

T g

f

 

   

 

denote texture feature vector of database image, then distance between them is given by:

2 1

Q T

d g g

Q T

i

g g

f

f

d

f

f

C. Shape Feature

(8)

International Journal of Emerging Technology and Advanced Engineering

Website: www.ijetae.com (ISSN 2250-2459,ISO 9001:2008 Certified Journal, Volume 4, Issue 11, November 2014)

218

The reason for choosing shape feature for describing an object is because of its inherent properties such as identifiability, affine invariance, and reliability and occlusion invariance, thus shape of the object has proved to be a promising feature based on which several image classification and retrieval operations can be performed [14].The shape descriptor are classified into two major kind namely Contour-based shape representation and description techniques and Region-based shape representation and description techniques. In the Counter-based shape techniques use only shape boundary information. In the Region based techniques it takes whole shape under consideration. There are some shape Descriptor methods like Geometric Moments Geometric Moments, Zernike Moments, Fourier Descriptor, Grid Method and Shape Matrix.

I. Fourier Descriptors

By applying Fourier transform on shape boundary Fourier Descriptor can be carried out. The co-efficient of Fourier transform are called Fourier descriptor of the shape. Shape boundary is a set of coordinates (xi, yi ), i = 1, 2, …, L, which are extracted out in the preprocessing stage by contour tracing technique [14]. The centroid distance

function is expressed by the distance of the boundary points from the centroid (xc, yc) of the shape

 

1

2 2 2

i i c i c

r

x

x

y

y

, i = 1, 2, …, L

Where xc, yc are averages of x coordinates and y

coordinates respectively. Due to the subtraction of centroid (which represents the position of the shape) from boundary coordinates, the centroid distance representation is invariant to shape translation [14].For applying Fourier transform normalized the boundary points of all the shape in database. The Fourier transform of ri , i=1,2,……N-1 is

given by [14]

1

0

1

2

exp

,

0,1,....,

1

N

n i

i

j

ni

u

r

n

N

N

N

The coefficients un, n = 0, 1, …, N-1, are usually called Fourier descriptors (FD) of the shape, denoted as FDn, n

=0, 1, …, N-1. The following feature vector are used as the Fourier descriptors to index the shape [14],

1 2 /2

0 0 0

,

,...,

N

FD

FD

FD

f

FD

FD

FD

 

Robustness is nice properties Fourier descriptor. It is being able to capture some perceptual characteristics of the shape and easy to derive. With Fourier descriptors, coarse shape features or global shape features and the finer shape features are captured by lower order coefficients and higher order coefficients respectively. For Fourier Descriptor Noise is not a problem because at very high frequencies noise are truncated out.

II.Css Descriptors

CSS descriptor is called Curvature scale Space. To obtain the CSS descriptor, first calculate the CSS counter map, the map is multi-scale organization of inflection points. To calculating CSS contour map, curvature is derived from shape boundary points (xi, yi), i = 1, 2, …, L :

2 2

32

i i i i i i i

k

x y

x y

x

y

Where

x

i,

y

i and

x

i,

y

iare the first and the second

derivatives at location i respectively [14].In the next step the shape is smoothed by applying Gaussian smooth:

' '

( , ),

( , )

i i i i

x

 

x

g i

y

 

y

g i

Where

means convolution, and g(i, σ) is Gaussian

function. As σ increases, the evolving shape becomes smoother and smoother [14]. Until no curvature zero-crossing points are found, continuous this process. CSS descriptors are constant in the translation. To achieve the Scale invariance it should be normalizing all the shapes into fixed number of boundary points. To achieve the Rotation invariance it should be circular shifting the highest peak to the origin of the CSS map. The similarity between two shapes is measured by the sum of the peak differences between all the matched peaks and the peak values of all the unmatched peaks [14]. The difficulties are arrived in the matching between two set of CSS descriptors are the dimensions of the two set of CSS descriptors. They are usually different and the CSS peaks of two similar shapes are usually not matching.

III. Zernike Moments

(9)

International Journal of Emerging Technology and Advanced Engineering

Website: www.ijetae.com (ISSN 2250-2459,ISO 9001:2008 Certified Journal, Volume 4, Issue 11, November 2014)

219

 

,

cos , sin

  

exp

nm nm nm

V

x y

V

 

R

jm

And

 

 

 2

2

0

( )! 1

! ! !

2 2

n m

s n s

nm

s

n s R

n m n m

s s s

   

 

     

 

   

   

   

Where n and m are subject to n-|m| = even, |m|≤n. Zernike polynomials are become complete when set of complex valued function added over the unit disk, i.e., x2 +

y2 = 1.Then the complex Zernike moments of order n with repetition m are defined as [14]:

* 2 2

1

( , )

( , ),

1

nm nm

x y

n

A

f x y V

x y x

y



Zernike moments are as same as to the Fourier transform, to expand a signal into series of orthogonal basis. The number of Moments truncated from the expansion is called the precision of shape representation. Basis functions of Zernike moments take the unit disk as their domain. Before calculating moments unit disk must be specified. Zernike moments descriptors do not need to know boundary information, making it suitable for more complex shape representation [14]. To overcomes the drawback of geometric moments in which higher order moments are difficult to construct, Zernike moments descriptors are constructed to arbitrary order.

IV. Grid Descriptors

In grid shape representation, a shape is projected onto a grid of fixed size, 16×16 grid cells for example [14]. If grid cells are assigned the value 1 then they are covered by the shape and if value 0 is assigned to grid cells then they are outside the shape. A shape number consisting of a binary sequence is created by scanning the grid in left-right and top bottom order, and this binary sequence is used as shape descriptors to index the shape [14]. To achieve scale, rotation and translation invariance for comparison of two shape using Grid Descriptor, several normalized process have to be done. To achieve this first find out the major axis. To achieve Rotation normalization the shape will be turning on so that the major axis is parallel with x-axis. To avoid multi normalization results for mirrored shape and flipped shape, the centroid of the rotated shape may be restricted to the lower-left part, or a mirror and a flip operation on the shape number are applied in the matching stage [14]. To achieve Scale normalization Shape has to be resized so that the length of the major axis is equal to the grid width, after that shift the shape to the upper-left of the grid.

In the set of Grid descriptor, the distance between two grid descriptors is simply the number of elements having different values. For example, the grid descriptors for the two shapes in Fig. 8 (a) and (b) are 001111000 011111111 111111111 111111111 111110011 001100011 and 001100000 011100000 111100000 111100000 011111100 000111000 respectively, and the distance between the two shapes will be 27 [14].

For example, the two shapes in Fig. 8 (c) and (d) are perceptually similar, but are very different under grid representation, for the major axis of shape (c) is horizontal while the major axis of shape (d) will be vertical [14].

(a) (b)

[image:9.612.330.556.271.424.2]

(c) (d)

Fig 8. Grid representation of shape[14]

III. CONCLUSION

In this paper some different types of methods are describe. All these methods given in paper are best to individual. These methods are used by other researchers to bring out different output. There are some combinations available of color and texture for retrieved image. Combinations of the three features have also available. But with the different techniques of color, texture and shape can get different and better result than previous. With this paper it is conclude that by using all different methods can get better and better results.

REFERENCES

[1] Hui Hui Wang, Dzulkifli Mohamad and N.A. Ismail ―Approaches, Challenges and Future Direction of Image Retrieval ‖ journal of computing, volume 2, issue 6, june 2010, issn 2151-9617,pp. 193-199.

[2] Ahmed J. Afifi and Wesam M. Ashour ―Content-Based Image Retrieval Using Invariant Color and Texture Features‖ 978-1-4673-2181-5/12/$31.00 ©2012 IEEE.

(10)

International Journal of Emerging Technology and Advanced Engineering

Website: www.ijetae.com (ISSN 2250-2459,ISO 9001:2008 Certified Journal, Volume 4, Issue 11, November 2014)

220

[4] Amandeep Khokher, Rajneesh Talwar, ―Content-based Image Retrieval:Feature Extraction Techniques and Applications‖ International Conference on Recent Advances and Future Trends in Information Technology (iRAFIT2012),pp.9-14.

[5] S. Mangijao Singh and K. Hemachandran, ―Content based Image Retrieval based on the Integration of Color Histogram, Color Moment and Gabor Texture‖ International Journal of Computer Applications (0975 – 8887) , Volume 59– No.17, December 2012. [6] P.S.Suhasini , Dr. K.Sri Rama Krishna, Dr. I. V. Murali Krishna

―Cbir Using Color Histogram Processing ‖ Journal of Theoretical andApplied Information Technology, Vol6. No1. pp116 122. [7] Aman Chadha,,Sushmit Mallik and Ravdeep Johar ―Comparative

Study and Optimization of Feature- Extraction Techniques for Content based Image Retrieval‖. International Journal of Computer Applications (0975 – 8887), Volume 52– No.20, August 2012 [8] Nadia Baaziz, Omar Abahmane and Rokia Missaoui, ―Texture

Feature Extraction In The Spatial-Frequency Domain For Content-Based Image Retrieval,‖

[9] Mangimala Singha and K.Hemachandran, ―Content Based Image Retrieval using Color and Texture,‖ Signal & Image Processing : An International Journal (SIPIJ) Vol.3, No.1, February 2012, pp.39-57 [10] Vibha Bhandari1, Sandeep B.Patil ,‖Comparison of CBIR

Techniques using DCT and FFT for Feature Vector Generation, ‖ International Journal of Emerging Trends & Technology in Computer Science (IJETTCS), Volume 1, Issue 4, November – December 2012, pp. 90-96.

[11] Ahmed J. Afifi andWesam M. Ashour, ―Image Retrieval Based on Content Using Color Feature,‖ International Scholarly Research Network, ISRN Computer Graphics, Volume 2012, Article ID 248285, 11 pages.

[12] Swapna Borde and Udhav Bhosle, PhD. ―Feature Vectors based CBIR in Spatial and Transform Domain,‖ International Journal of Computer Applications (0975 – 8887), Volume 60-No.19, December 2012, pp. 34-45.

[13] Shereena V.B.1and Julie M. David2 ―Content Based Image Retrieval : A Review‖ pp. 65-77.

[14] Sai Anand.C, Tamilarasan.M and Arjun.P ―A Study on Curvature Scale Space ‖ International Journal of Innovative Research in Computer and Communication Engineering,Vol. 2, Special Issue 3, July 2014,pp.168-174.

[15] Dr. Fuhui Long, Dr. Hongjiang Zhang and Prof. David Dagan Feng, ―Fundamentals Of Content Based Image Retrieval ‖

[16] Mohamed A. Helala, Mazen M. Selim, and Hala H. Zayed ―A Content Based Image Retrieval Approach Based On Principal ‖ IJCSI International Journal of Computer Science Issues, Vol. 9, Issue 4, No 1, July 2012,pp.204-213.

[17] Mas Rina Mustaffa, Fatimah Ahmad, Rahmita Wirza O.K. Rahmat, Ramlan Mahmod ―Content-Based ImageRetrieval Based On Color-Spatial Features ‖, Malaysian Journal of Computer Science, Vol. 21(1), 2008, pp. 1-12

[18] Greg Pass, Ramin Zabih ―Histogram Re_nement for Content-Based Image Retrieval‖

[19] Abhinav Deshpande, and S.K.Tadse ―Design Approach for Content-Based Image Retrieval Using Gabor-Zernike Features‖. International Archive of Applied Sciences and Technology. Volume 3 [2] June 2012: 42 – 46 ISSN: 0976-4828.

[20] Hamid A. Jalab ―Image Retrieval System Based on Color Layout Descriptor and Gabor Filters,‖ 2011 IEEE Conference on Open Systems (ICOS2011), September 25 - 28, 2011, Langkawi, Malaysia, pp. 32-36.

[21] Tomasz Andrysiak and Michał Chora´S,‖Image Retrieval Based On Hierarchical Gabor Filters, ‖ Int. J. Appl. Math. Comput. Sci., 2005, Vol. 15, No. 4, 471–480.

[22] Fatma Chaker, Faouzi Ghorbel and Mohamed Tarak Bannour ― Content-Based Shape Retrieval Using Different Affine Shape Descriptors,‖ VISAPP 2008 - International Conference on Computer Vision Theory and Applications, pp. 497-500.

[23] Dengsheng Zhang and Guojun Lu,‖Content-Based Shape Retrieval Using Different Shape Descriptors : A Comparative Study,‖. [24] Sami Brandt, Jorma Laaksonen And Erkki Oja,‖Statistical Shape

Features for Content-Based Image Retrieval,‖ Journal of Mathematical Imaging and Vision 17: 187–198, 2002.

Figure

Fig. 3. A two-level wavelet decomposition [8].
Fig. 5. A set of real impulse responses: multiscale, multi-orientation Gabor wavelet filters [8]
Fig 8. Grid representation of shape[14]

References

Related documents

[4] S.P.Rajaram et al., “Optimal Placement of Distributed Generation for Voltage Stability Improvement and Loss Reduction in Distribution Network”

Figure 3.2: Graphical representation of dependency structure tree for the sentence: Please refrain from making personal attacks .. generate a rule for detecting this type of

As the Rule Verification in Creating module detects the conflicts of rules on device, the Creating Conflict Detection (CCD) compares the rule with the other related rules in

(Hare's account does not fit.. into this pattern, but will be analysed in detail later in this chapter.) Another strand of the tradition tries to avoid the obvious problems

Patients with stroke may suffer several types of pain, including articular pain, musculoskeletal pain, painful spasticity, headache, and neuropathic central post-stroke

An anomalous diffusion model based on a new general fractional operator with the Mittag Leffler function of Wiman type Liang et al Advances in Difference Equations (2018) 2018 25

Adaptive tracking control for a class of uncertain nonlinear systems with infinite number of actuator failures using neural networks Lv and Wang Advances in Difference Equations (2017)